JPH02263281A - Parallel picture processor - Google Patents

Abstract

PURPOSE:To enable exact edge detection without generating a refractory point to an edge by two-dimensionally allocating the plural couples of light receptive elements, for which the respective couples are divided into two areas such as an area to generate a positive output and an area to generate a negative output when light is made incident thereon, to make the absolute value of the sum of the outputs from the two areas into an output to respective threshold elements. CONSTITUTION:To edge detecting cells to be the respective threshold elements, a unit receptive area 13 composed of plural photo-electric converting cells, for example, four photo-electric converting cells A+, A, B+ and B is allocated. These photoelectric converting cells A+, A, B+ and B are rectangularly arranged and the respective couples are divided into the two areas such as the areas A+ and B+ to generate the positive output and the areas A and B to generate the negative output when the light is made incident thereon. Then, the absolute value of the sum of the outputs from the two areas is made into the output of the light receptive element in each couple. Thus, the edge detection can be executed without generating the refractory point to the edge.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to pattern recognition of graphical information such as characters and symbols, particularly edge information detection, and can be further applied to artificial visual devices for robot vision. This invention relates to a parallel image processing device.

In conventional pattern recognition of technical graphic information, it is necessary to recognize its characteristics. For such recognition, we focused on the point that it would be effective to elucidate the feature extraction mechanism of the living body, and we devised a parallel image processing device that utilizes the multilayered structure of neurons in the visual nervous system of the living body. In other words, a light-receiving layer has a two-dimensional arrangement of a large number of photoreceptive elements that receive input images, and a threshold has a two-dimensional arrangement of a large number of threshold elements that receive output from each photoreceptor. a layer of elements, generating from each threshold element an output equal to a function of a weighted sum of the received inputs;
A parallel image processing device that forms an edge information output corresponding to an input image is modeled after the visual system of a living body.

Regarding such image processing, for example,
There is one shown in Publication No. 4901.

Here, an example of the configuration of such a conventional parallel image processing device, particularly a threshold element having an edge detection function, will be described. In this case, the threshold element receives and inputs outputs from a plurality of photoreceptor elements.In the following explanation, a photoelectric conversion cell such as a photodiode is used as the photoreceptor element, and an edge cell is used as the threshold element. Detection cells shall be used.

First, FIG. 5 shows a photoelectric conversion cell layer l as a light-receiving layer. This photoelectric conversion cell layer 1 has a large number of photoelectric conversion cells (photoreceptor elements) 2 arranged in a two-dimensional matrix on the imaging plane of the photographing lens 3 for the input image. Therefore, an image corresponding to the input image is formed on each photoelectric conversion cell layer l, and each photoelectric conversion cell 2 converts it into an electrical signal corresponding to the light intensity at the arranged two-dimensional coordinates and outputs it. I will put it out. Here, it is assumed that the output of each photoelectric conversion cell 2 can be handled independently.

Further, FIG. 6 shows an edge detection cell layer (threshold element layer) 4 that is combined with such a photoelectric conversion cell layer 1. This edge detection cell layer 4 has a large number of edge detection cells (threshold elements) 5 arranged in a two-dimensional matrix. Each edge detection cell 5 receives output signals from the plurality of photoelectric conversion cells 2 and outputs a signal equal to a certain function of a weighted sum of these received inputs.

Here, for example, as extracted and shown in FIG. 7, each edge detection cell 5 is combined so as to receive the output signals of nine photoelectric conversion cells 2 forming a unit receiving area 6 in a 3×3 matrix format. ing. Now, the nine photoelectric conversion cells 2 constituting the unit receptor area 6 are defined as PD, , PD,
,,PD,,,PD,,.

If shown as PD, , PD, , PD, , PD, , PD, , PD, , shown with diagonal lines in FIGS. 6 to 9, is in the unit receiving area 6. The remaining PD, , ~PD, , PD, , ~P
D, , are the surrounding area cells. Furthermore, such unit receiving areas 6 overlap with adjacent unit receiving areas 6.

Here, the cells in the central region send a signal of positive potential (excitement) to the edge detection cell corresponding to the time of light reception (when light is incident), and the cells of the peripheral region send a signal of negative potential (inhibition) to the edge detection cell corresponding to the time of light reception. ) signals.

Therefore, considering a signal processing system for 1942 minutes, as shown in Fig. 8, peripheral area cells PD,..., PD s
After the outputs of a are added by an adder 7, they are added to an inverter 8.
By passing through the central region cells P D
, , and will be input to the corresponding edge detection cell 5.

An edge detection cell (threshold element) having such input characteristics is called a threshold element having an ON-centered receptive field. Figure 9 shows 3X
3 shows ONN-centered receptive force mode processing in the case of unit receptive area 6 with 3=9 photoelectric conversion cells PD, . . . ~PD s m.

The processing when the unit receiving area 6 has a 3×3 configuration will be explained mathematically. First, the output of each photoelectric conversion cell PD, , ~PD, , is Uol, ~U''11, and these outputs U°,.

~U0. ．． Assuming that the weighting coefficients (weighting coefficients) at the time of input are respectively CI l to Ca 6, the input lN1j to the edge detection cell 5 is as follows: IN'ij = U' + + C+ + + U' + m
C+ m+...+U e, Cs m+U'''□C,
,+U@,,C,.

becomes. Here, regarding the weighting coefficients CI l to C0,
Due to the ON-centered type, C,, =C,, =C,, =C,, =C,, =C,, =C
,o=c,,=ch<.

C,,=Ce)0. Further, the relationship between ch and Ce is set as 18Chl = lCel. As a result, the output U'fj from the edge detection cell 5 is output according to the following function.

however, e=U”,,C,, h=U',,C,,+U'',,C,,+U'',,C,.

10U”,,C:,,+U”,,C,,+U’,,C,.

+U”,,c,,+u”,,c,.

shall be.

Then, the image edge exists within the unit receiving area 6 of 3×3=9 cells, and the amount of light incident on the central region cell PD, , U°8. and other peripheral area cells PD,
The amount of light U incident on ~PD, , PD, , ~PD, ,
°1. 〜U″Ill U″ll〜U°,, Which ratio is 1
:8, the edge detection cell 5 corresponding to the unit receiving area 6 outputs U'ij. In this way, the edge detection cells 5 have edge detection ability.

This point will be further explained with reference to FIG. One edge detection cell 5 is connected to nine photoelectric conversion cells PD, . The evaluation can be made based on the output when the light shielding plate 9 corresponding to the image is displaced from left to right in the X direction to sequentially increase the area of light shielding of the unit receiving area 6. Here, the displacement of the light shielding plate 9 represents the coordinates of the boundary of light shielding, and the changes in the value of e-h (relative value) and the value of 1e-hl (relative value) related to the output U'ij are shown. , respectively, as shown in FIGS. First, the value of 1e-hl is It increases in proportion to the shaded area. Furthermore, the light shielding progresses, and the central photoelectric conversion cell PD, PD,...

When the column PD, , is shaded, the value of e, ie, the decrease in the output from the central region cell PD, , becomes dominant.

When these photoelectric conversion cells PD, , PD, , PD, , are shielded from light by exactly 1/2, the value of 1e-hl becomes O. And the photoelectric conversion cell PD in the center.

When the PD,...,PDo rows are all shielded, 1e-h
The value of l is the remaining charge conversion cell PD, , PD, , .

Determined by the output value from the P D, column. As the light shielding progresses further, the photoelectric conversion cells PD on the right... PD...

When the PD, , column is shaded, the value of 1e-hl becomes smaller in proportion to the shaded area.

The problem to be solved by the invention is that in the conventional edge detection method, among the plurality of photoreceptor elements input to each threshold element, the light incident signal from the central element is used as a positive signal, and the light incident signal from the peripheral element is used as a positive signal. It constitutes an ON-centered receptive field in which the incident signal is a negative signal. Then, after weighting the center signal and the peripheral signal, the difference is taken and the absolute value is used as the edge output. However, in this case, when the edge passes through the receptive field (receptive area), the first
As can be seen from the diagram O, the incidence of light at the center and the periphery becomes equal, resulting in an edge refractory point where edges cannot be detected.

Means for Solving the Problems A light-receiving layer has a two-dimensional arrangement of a large number of photo-receptive elements that receive input images, and a large number of threshold elements that receive outputs from the plurality of photo-receptive elements are two-dimensionally arranged. a parallel image processing device comprising an array of threshold element layers, each threshold element generating an output equal to a function of a weighted sum of received inputs, and forming an edge information output corresponding to the input image; A two-dimensional photoreceptor element is divided into two regions, each group of which produces a positive output and a region which produces a negative output when light enters the element, and the output is the absolute value of the sum of the outputs of the two regions. Multiple sets were assigned to

A plurality of sets of light-receiving elements are two-dimensionally assigned to each operational threshold element, and each set is divided into two regions: a region that produces a positive output when light is incident, and a region that produces a negative output, The absolute value of the sum of the outputs of the two regions is used as the output of each set of photoreceptor elements, and if these sums are used as the input to the threshold element, edge detection can be performed without creating a refractory point for edges. It becomes possible.

Embodiment An embodiment of the present invention will be described with reference to FIGS. 1 to 3. This embodiment is based on the edge detection cell method described above, and the same parts as those shown in FIGS. 5 to 10 are indicated by the same reference numerals.

In this embodiment, for each threshold element, which is an edge detection cell, a unit receptor area 1 consisting of a plurality of photoelectric conversion cells, for example, four photoelectric conversion cells A, , i, B, , B- as shown in FIG.
3 is assigned. These photoelectric conversion cells A...
A-, B, , B- are arranged in a rectangular shape as shown in FIG. Here, the photoelectric conversion cells A, , A- form a certain group A, and are arranged diagonally, and the photoelectric conversion cells B, , B- also form a certain group B, and are arranged diagonally. placed in position. In addition, photoelectric conversion cells A, B indicate areas that produce positive output when light is incident, and photoelectric conversion cells A-1B
- indicates a region that produces a negative output when light is incident. Therefore, A
Both the group and the B group consist of two regions that produce positive and negative outputs, respectively.

Under such a configuration of the unit receptor area 13, the signal from group A is the absolute value of the sum of the signal from photoelectric conversion cell A and the signal from charge conversion cell A-. In addition, the signal from group B also includes the signal from photoelectric conversion cell B4 and the signal from photoelectric conversion cell B
It is considered to be the absolute value of the sum of the signals from -. The input signal E to the edge detection cell (threshold element) corresponding to such a unit receptor area 13 is the sum of the outputs of each set, that is, E=lA, -A-1+lB, -B -1.

According to such a configuration, when the light shielding plate (edge) 9 moves from left to right in the X direction as shown in FIG. 2(a), an edge signal E as shown in FIG. 2(b) is generated. becomes. That is, an edge detection refractory point such as E=0 does not occur in the middle, and accurate edge detection is possible.

This is the same even when the light shielding plate 9 moves from right to left, from top to bottom, and from bottom to top.

Furthermore, as shown in FIG. 3(a), even when the light shielding plate 9 crosses the unit receiving area 13 in an oblique direction, the edge detection ability is maintained without producing an edge refractory point. However, in this case.

The intensity (sensitivity) of the signal E as shown in FIG. 2(b) decreases. However, if the fact that the sensitivity differs depending on the direction of edge movement is actively utilized, the sensitivity can also have selectivity in the direction of movement.

The configuration of the unit receiving area for one threshold element is not limited to the example illustrated in FIG. 1, but may be the number and arrangement as illustrated in FIG. 4, for example.

First, Figure (a) shows four photoelectric conversion cells A, , A-.

B, , B- are arranged at a distance. The same figure (b
) is one in which four photoelectric conversion cells A, , A-, B, and B- are arranged in a cross shape. In these cases as well, there is no edge refractory point and the sensitivity has orientation selectivity.

However, the azimuth with respect to sensitivity differs by 45 degrees between FIG.

In addition, FIG. 12(c) shows an example of a unit receptor area configuration with suppressed azimuth selectivity of sensitivity, that is, the pattern of FIG.
-, B, and B- correspond) and the pattern in (b) of the same figure (charge conversion cells αφ, α-2β, and β located in the middle)
- is equivalent). This cancels out differences in sensitivity to orientation. The edge signal E in this case becomes E=lA, -A-l+lB, -B-1 +1α, -α-1+1β, and -β-.

Effects of the Invention As described above, in the present invention, for each threshold element, each set is divided into two regions, a region that produces a positive output when light is incident, and a region that produces a negative output, and the sum of the outputs of the two regions is divided into two regions. Since multiple sets of photoreceptor elements whose output is absolute value are allocated two-dimensionally, the sum of the outputs of each set is input to the threshold element, allowing accurate edge detection without creating refractory points for edges. can be made possible.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram schematically showing the arrangement of the unit receiving area in an embodiment of the present invention, FIGS. 2 and 3 are explanatory diagrams showing the edge detection operation, and FIG. An explanatory diagram schematically showing a modification of the arrangement configuration, FIG. 5 is a perspective view of the principle of a photoelectric conversion cell layer showing a conventional example, FIG. 6 is a perspective view of the principle of an edge detection cell layer, and FIG. 7 is one of them. FIG. 8 is a block diagram, FIG. 9 is an explanatory diagram of the ONN-centered receptive force method, and FIG. 10 is an explanatory diagram of edge detection ability. 1... Light-receiving layer, 2... Photo-receiving element, 4... Threshold element layer, 5... Threshold element, 13... Unit acceptance area, A, B
Ya, α,. β,... Positive output area, i, B-, α-9β-... Negative output area Applicant Rico Co., Ltd. 1" Figure 3
7 J, Figure 3 (a) is Figure (b) 3 (Re~Atsushi Figure 5 JJO

Claims (1)

[Claims] a light-receiving layer in which a large number of light-receiving elements for receiving input images are two-dimensionally arranged; and a threshold element layer in which a large number of threshold elements for receiving outputs from the plurality of light-receiving elements are two-dimensionally arranged; , wherein each threshold element generates an output equal to a function of a weighted sum of received inputs to form an edge information output corresponding to the input image; is divided into two regions, a region that produces a positive output and a region that produces a negative output when light is incident, and two-dimensionally allocates a plurality of sets of photoreceptor elements whose output is the absolute value of the sum of the outputs of the two regions. Features of parallel image processing device.